The Interpretation of Pyrite Laser Ablation-Inductively Coupled Plasma-Mass Spectrometry Maps Using Machine Learning: A Case Study of the Colosseum Au Deposit, Southern California

Abstract

This study explores the application of machine learning techniques for an enhanced interpretation of pyrite laser ablation-inductively coupled plasma-mass spectrometry (LA-ICP-MS) maps. The Colosseum Au deposit, in southern California, was considered as a case study. Colosseum is genetically related to a rhyolitic breccia-pipe complex, where Au mineralization is associated with two main pyrite generations—early pyrite and late pyrite. Our machine learning workflow involves the detection of distinct compositional zones in individual maps through unsupervised clustering, and a second clustering step where these zones are grouped by compositional similarity, enabling the direct comparison between different maps and providing a compositional overview of pyrite representative of the various styles of mineralization present in the deposit. Clustering of individual maps correctly differentiated between distinct growth zones in early pyrite, fractures that crosscut early pyrite growth, and zones of late pyrite growth, matching petrographic observation. All the zones detected by this first step, in turn, were classified into two compositionally distinct groups and a third transitional group, enabling the direct comparison between maps while keeping petrographic consistency. For Colosseum, our approach revealed that (1) Au is more abundant in late pyrite than early pyrite, but significant amounts can be found in both generations and in both Colosseum mineralized breccia pipes; (2) the transition from early to late pyrite is represented by a change from a Co-Ni-Te–rich end member to a Cu-Ag-Zn-Sb-Tl–rich end member; and (3) Au is directly correlated with As in both pyrite generations.